Re: Oracle 8i (8.1.7.0.1) + Redhat Linux 7.2 = Cannot create tablespace file > 2 gb

"Howard J. Rogers" wrote:
>
> "Sean M" <smckeown_at_earthlink.net> wrote ...
>
> > In practice, most incomplete recoveries of production databases roll
> > back to a point in time earlier than the data contained in the online
> > redos. That's been my experience, and I bet most others' that do this
> > on real production systems. Why? Because most real production systems
> > switch logs rather frequently (say, a few times per hour),
>
> I hope not. But perhaps things are different south of the equator. Most DBAs
> I know of here would switch every hour or so. And I disapprove (since they
> are only protecting themselves against Instance failure by doing so, and I
> happen to think that hardware and most O/Ses, not to mention UPSes have
> gotten to the point where I hope we can start to stop worrying about
> instance failures).

[What the hell, more thread drift...]

I'd postulate at least 2 other reasons for more frequent log switches than once a day. First, for many of our databases (and in many other large shops out there), we generate 50-200 GB of redo a day. Making 50-200 GB online redo logs really isn't very practical for backup and recovery, for many of the same reasons we've seen in this thread for not using large datafiles (i.e. higher granularity of the recovery process is usually beneficial). Secondly, even if we did only switch once a day for these databases, that would incur one BIG checkpoint (by big I mean a LOT of dirty buffers being written to datafiles, all at once). That means once a day we get a HUGE performance spike, as opposed to many smaller spikes (a few times per hour). For 24/7 systems, we can't really tell our users that everything runs great except for that one 45-minute period where you can't do a darn thing because, sorry, we're checkpointing. But we can tell them that everything runs pretty well all day long.

And I wouldn't understate the frequency of instance crashes. If they weren't still a problem with modern hardware we wouldn't see such an interest in RAC. Bugs, hung processes, memory leaks, cpu failures, you name it - instance failure isn't going away anytime soon. I think you're once-a-day switch idea will be a hard sell in production, but don't let me stop you. :)

Regards,
Sean Received on Sat Jun 29 2002 - 02:06:18 CDT